Alpha Beta Filter Dashboard

Filecoin implements a Alpha Beta Filter for estimating the Block Reward and the QA Network Power on the Sector Initial Pledge Calculations.

• Sector Pledge = Storage Pledge + Consensus Pledge
• Initial Storage Pledge = Estimate of 20 Days of Block Reward
• Initial Consensus Pledge = 30% Total Circulating Supply Sector QA Power / max(Baseline Function , Estimate of QA Network Power)

Current parameters are:

• $\alpha=9.25e-4$
• $\beta= 2.84e-7$ per epoch

References

Source Code

• Alpha Beta Filter: https://github.com/filecoin-project/specs-actors/blob/f6a8d2a6f1573b2786cd573a2ea005aa04b1eb56/actors/util/smoothing/alpha_beta_filter.go#L78
• Reward Actor: https://github.com/filecoin-project/specs-actors/blob/f6a8d2a6f1573b2786cd573a2ea005aa04b1eb56/actors/builtin/reward/reward_state.go
• Power Actor: https://github.com/filecoin-project/specs-actors/blob/f6a8d2a6f1573b2786cd573a2ea005aa04b1eb56/actors/builtin/power/power_state.go#L298

Setting-up

%load_ext autotime
%load_ext autoreload
%autoreload 2

time: 10.8 ms (started: 2021-09-19 00:10:26 +00:00)

# External depences
import pandas as pd
import numpy as np
import plotly.express as px

# Move path to parent folder
import sys
sys.path.insert(1, '../')

# Internal dependences
from filecoin_metrics.connection import get_connection, get_connection_string
from filecoin_metrics.metrics import *

time: 589 ms (started: 2021-09-19 00:10:26 +00:00)

conn_string = get_connection_string('../config/sentinel-conn-string.txt')
connection = get_connection(conn_string)
BASELINE_CROSSING = '2021-04-02 05:00+00'

time: 237 ms (started: 2021-09-19 00:10:26 +00:00)

Block Reward Estimator

QUERY = f"""
/* 
Retrieve block reward & qa network power daily averages / stdevs for
- Measurements
- Position Estimates
- Velocity Estimates
*/

select
    /* Timestamp */
	date_trunc('HOUR', to_timestamp(height_to_unix(cr.height))) as timestamp,

    /* Reward measurements & estimates */
    avg(cr.new_reward::numeric / 10^18) as reward_avg,
    stddev(cr.new_reward::numeric / 10^18) as reward_std,
	avg(cr.new_reward_smoothed_position_estimate::numeric / (2^128 * 10^18)) as reward_position_avg, /* FIL */
    stddev(cr.new_reward_smoothed_position_estimate::numeric / (2^128 * 10^18)) as reward_position_std, /* FIL */
    avg(cr.new_reward_smoothed_velocity_estimate::numeric / (2^128 * 10^18)) as reward_velocity_avg, /* FIL per epoch */
    stddev(cr.new_reward_smoothed_velocity_estimate::numeric / (2^128 * 10^18)) as reward_velocity_std, /* FIL per epoch */

    /* Power measurements & estimates */
    avg(cp.total_qa_bytes_power::numeric / 1024^5) as qa_power_avg, /* PiB */
    stddev(cp.total_qa_bytes_power::numeric / 1024^5) as qa_power_std, /* PiB */
	avg(cp.qa_smoothed_position_estimate::numeric / (2^128 * 1024^5)) as qa_power_position_avg, /* PiB */
    stddev(cp.qa_smoothed_position_estimate::numeric / (2^128 * 1024^5)) as qa_power_position_std, /* PiB */
    avg(cp.qa_smoothed_velocity_estimate::numeric / (2^128 * 1024^5)) as qa_power_velocity_avg, /* PiB per epoch */
    stddev(cp.qa_smoothed_velocity_estimate::numeric / (2^128 * 1024^5)) as qa_power_velocity_std /* PiB per epoch */
FROM chain_rewards cr
join chain_powers cp on cp.height = cr.height
where cr.height > 148888 /* Mainnet Launch Block Height */
group by timestamp
order by timestamp
"""

query_df = (pd.read_sql(QUERY, connection))

time: 2.53 s (started: 2021-09-19 00:10:26 +00:00)

fig = px.line(query_df, 
            x='timestamp',          
            y=['reward_avg', 'reward_position_avg'],
            title='Hourly Averages of Block Reward')
fig.show()

time: 1.08 s (started: 2021-09-19 00:10:29 +00:00)

fig = px.line(query_df, 
            x='timestamp',          
            y=['reward_std', 'reward_position_std'],
            log_y=True,
            title='Hourly Standard Deviation of Block Rewards')
fig.show()

time: 752 ms (started: 2021-09-19 00:10:30 +00:00)

velocity_z = (query_df.reward_velocity_std / query_df.reward_velocity_avg).abs()

fig = px.line(query_df.assign(std_per_avg=velocity_z), 
            x='timestamp',          
            y='std_per_avg',
            log_y=True,
            title='Block Reward Velocity Hourly Standard Score')
fig.show()

time: 398 ms (started: 2021-09-19 00:10:31 +00:00)

relative_error = (query_df.reward_avg - query_df.reward_position_avg) / query_df.reward_avg
df = query_df.assign(relative_reward_error=relative_error)
fig = px.line(df,
            x='timestamp',
            y='relative_reward_error',
            title='Mean Relative Error on Hourly Averages of Block Rewards')
fig.update_layout(yaxis=dict(tickformat=".2%"))
fig.show()

time: 392 ms (started: 2021-09-19 00:10:31 +00:00)

QA Network Power Estimator

fig = px.line(query_df, 
            x='timestamp',          
            y=['qa_power_avg', 'qa_power_position_avg'],
            title='Hourly  Averages of QA Power')
fig.show()

time: 766 ms (started: 2021-09-19 00:10:32 +00:00)

fig = px.line(query_df, 
            x='timestamp',          
            y=['qa_power_std', 'qa_power_position_std'],
            log_y=True,
            title='Hourly Standard Deviation of QA Power')
fig.show()

time: 869 ms (started: 2021-09-19 00:10:33 +00:00)

relative_error = (query_df.qa_power_avg - query_df.qa_power_position_avg) / query_df.qa_power_avg
df = query_df.assign(relative_reward_error=relative_error)
fig = px.line(df,
            x='timestamp',
            y='relative_reward_error',
            title='Mean Relative Error on Hourly Averages of QA Power')
fig.update_layout(yaxis=dict(tickformat=".1%"))
fig.show()

time: 399 ms (started: 2021-09-19 00:10:33 +00:00)

velocity_z = (query_df.qa_power_velocity_std / query_df.qa_power_velocity_avg).abs()

fig = px.line(query_df.assign(std_per_avg=velocity_z), 
            x='timestamp',          
            y='std_per_avg',
            log_y=True,
            title='QA Power Velocity Hourly Standard Score')
fig.show()

time: 391 ms (started: 2021-09-19 00:10:34 +00:00)